Crypto Analysis MCP Server providing cryptocurrency price data and technical analysis indicators (SMA, RSI, MACD) using CCXT and TA-Lib, with support for A-stock, HK-stock, US-stock, and ETF markets.
This MCP server exhibits significant definition quality gaps across multiple dimensions. While tool names follow verb_noun conventions and all tools have descriptions, the descriptions are in Chinese and critically underdocumented. Input schemas are present but lack parameter-level descriptions in most tools. Output schemas are not documented. The server handles financial data (stocks, crypto, ETFs) with no error handling guidance, missing permission gates, and no audit trail patterns. Average description length ~30 chars (below the 194 baseline), and parameter descriptions are minimal or absent. The 'ping' tool is vestigial (removed from MCP spec). Tool composition is reasonable but lacks reference chaining documentation.
计算MACD指标
计算相对强弱指数(RSI)
计算简单移动平均线(SMA)
获取A股K线数据
获取A股当前价格
获取ETF K线数据
获取ETF当前价格
All tool descriptions are in Chinese (e.g., '获取A股K线数据') and critically short (<30 chars). Descriptions must be in English and between 10-1024 characters, with explicit WHAT/WHEN/HOW guidance for LLM tool selection. Current descriptions provide minimal context for an English-speaking LLM to determine when or why to select these tools.
Input parameter descriptions are missing or extremely minimal. For example, 'timeframe' parameter has description '时间框架' (Chinese for 'time framework'), not explaining valid values like '1h', '1d', or valid ranges. Per pattern baseline, 100% of A+ tools have parameter descriptions, and LLMs cannot infer meaning from names alone (e.g., 'limit' could mean result count, time limit, or API call limit).
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 53 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 0 | - | v1 |
获取美股K线数据
获取美股当前价格
Server health check
Output schemas are not documented. LLMs need to know what fields they will receive (e.g., from get_a_stock_candles: does it return 'open', 'high', 'low', 'close', 'volume'? as numbers or strings? with what timezone?). Without documented output, LLMs cannot plan downstream tool calls or extract the right data for users.
No error handling guidance provided. Tools fetch external financial data (via akshare, ccxt, yfinance) which can fail due to network issues, invalid symbols, rate limits, or API changes. Error responses must tell LLMs what to do next (retry, search for valid symbol, etc.). Current code has no recovery guides.
Parameters lack constraints and ranges. 'limit' and 'history_len' are unbounded integers, LLMs could pass limit=10000 or history_len=999999, causing timeouts or API errors.
'timeframe' and 'period' parameters should be enums, not free-form strings. For example, get_a_stock_candles accepts 'timeframe', but are valid values '1h'/'1H'/'1min'/'1min'/'daily'/'D'? Without enums, LLMs hallucinate invalid values. Specify: timeframe: enum [1min, 5min, 15min, 30min, 1h, 1d, 1w, 1M].
'adjust' parameter in stock tools accepts 'qfq' (forward adjustment), 'hfq' (backward adjustment), or presumably none, but only Chinese description provided. Describe in English with enum: adjust: enum ['qfq', 'hfq', 'none'] (default 'qfq' for forward-adjusted prices).
'symbol' parameter description is bare and in Chinese. For stock tools: 'A股代码,6位数字' (A-share code, 6-digit number), but is it always numeric? Can it include leading zeros? What happens if the symbol doesn't exist? Improve: 'A-share code: 6-digit identifier (e.g., 000001 for SZZS). Invalid symbols return a 'symbol not found' error; use search_stock_symbols() to discover valid codes.' But search_stock_symbols() doesn't exist, so add it or document valid symbol format more precisely.
No audit trail or logging for financial data access. This server handles market data fetches; even though read-only, production deployments should log WHO called what tool, WHEN, and with which parameters, for compliance and debugging. No scope declarations (read:market_data, read:crypto) are visible.
The 'ping' tool is a vestigial pattern. The MCP spec (2026-07-28) removed ping, servers no longer need to implement it. This tool adds noise and wastes LLM reasoning cycles. Remove it.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are visible. While all tools are marked Risk: READ_ONLY in the assessment, the MCP spec allows explicit tool annotations to declare these semantics to clients. Adding toolAnnotations would improve spec alignment.